Recognition Technology Of Pine Wood Affected By Pine Wilt Disease Based On YOLOv5

It is very important for the prevention and control of pine wilt disease(PWD)to effectively recognise the affected trees.Different from the traditional recognition method of infected pine wood,the recognition method based on deep learning has lower cost of feature data processing,faster speed and higher accuracy.In order to solve the problems such as the selection of image acquisition height,in the recognition process of pine wood affected by PWD,such as dense objects,diverse scales,complex background,and image color easily affected by external environment.In this paper,the dataset of pine wood affected by PWD was established and the recognition algorithm based on YOLOv5 was studied and improved to help forestry personnel determine an appropriate image acquisition height of infected pine woodand recognize PWD effectively.In view of the above research contents,the main work of this paper is as follows:(1)Establishment of dataset of pine wood affected by PWD.We used an unmanned aerial vehicle(UAV)to collect images of the infected pine wood at three different heights of 200/220 m,100m and 60/80 m and labeled the images of the selected pine wood affected by PWD.In order to increase the diversity of the input images,ensure the robustness of YOOv5 and subsequent improved algorithm to images from different environments,the images at different heights in the dataset are mirrored,changed in Gaussian noise,changed in color brightness,changed in contrast and flipped to perform image amplification and preprocessing to establish the SYLG-Pine Wilt Disease Dataset(SYLG-PWD Dataset)of pine wood affected by PWD.(2)In view of the problem that in the process of recognition of infected pine wood,such as the determination of appropriate image acquisition height,complex background,image color vulnerable to light and other external environments,Combining Moasic data enhancement,Focus unit,spatial pyramid pooling unit and color attention mechanism C3C＿2(Convolution3Coordinate＿2)module based on Coordinate Attention,a recognition algorithm CA-YOLOv5 of pine wood affected by PWD based on YOLOv5 was proposed.The experimental results show that the mean average precision(m AP)of the dataset at 100 m acquisition height is the highest.Compared with Faster R-CNN algorithm,the m AP value of YOLOv5 is 16.8% higher and the detection speed increased by 34.1 frames per second.So,we help forestry personnel to determine a suitable image acquisition height of infected pine wood.What’s more,CA-YOLOv5 algorithm achieves an m AP value of 88.1%,which is 17.9%,1.4% and 3.4% higher than Faster R-CNN,YOLOv3 and YOLOv5 algorithm respectively.Compared with Faster R-CNN and YOLOv3 algorithms,the model size is reduced by 87% and 65.1%respectively.(3)In view of the problem of dense targets and multiple scales in the process of image acquisition of damaged trees,a recognition algorithm CIGE-YOLOv5 of pine wood affected by PWD based on CA-YOLOv5 was proposed.First,combining genetic mutation operation,Io U(Intersection over Union)and K-means,an anchor frame generation algorithm IG-K-means was proposed.Then,IG-K-means algorithm and CA-YOLOv5 algorithm were combined to generate CIG-YOLOv5 algorithm for recognizing infected pine wood of different scales.On this basis,a new infected pine wood recognition algorithm CIGE-YOLOv5 based on EIo U(Efficient Intersection over Union)and CIG-YOLOv5 was proposed to improve the recognition accuracy of dense infected pine wood.The experimental results show that the m AP value of CIGE-YOLOv5 algorithm is 19.1%,2.6%,14.1% and 1.2% higher than that of Faster RCNN,YOLOv3,Varifoval Net and CA-YOLOv5 respectively,achieving 40.2 FPS real-time detection speed.